Convolutional Neural Networks (Part 1)
Channel:
Subscribers:
5,330
Published on ● Video Link: https://www.youtube.com/watch?v=Aptm1TqyWPA
1Deep Learning tutorial series.
Notebook and Slides can be found at: https://github.com/tanchongmin/TensorFlow-Implementations
A small correction:
The strides (1, 1) parameter. The first parameter should be the height, and the second parameter should be the width, according to TensorFlow documentation.
On Filter weights:
If you want the 'X' or the triangle shape to be distinct from the background, apart from having weights 1 at the shape itself, we should also put perhaps -1 weights (instead of weights being 0) on the parts that are not the shape, so that we penalize the input for having high intensity pixels on the parts that are not the shape of interest. This is something that would make the filter select a sharper feature.